qwen
Qwen: Qwen3 235B A22B Thinking 2507
qwen/qwen3-235b-a22b-thinking-2507
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
Qwen3-235B-A22B-Thinking-2507 is an open-weight, Apache-2.0 Mixture-of-Experts text model from Qwen with 235B total parameters and 22B activated per token, designed to run in thinking mode only. The vendor reports state-of-the-art results among open thinking models on math, coding, and agentic benchmarks; those figures are self-reported. It natively supports a 262K-token context with a documented path toward 1M tokens. It is a plausible fit for complex reasoning and tool-calling workloads, subject to validation on your own tasks.[1][2]
Strengths
- Vendor-published evaluations report strong results on reasoning, math, coding, multilingual, and agentic benchmarks versus the prior Qwen3 thinking release.[2]
- Native 262K-token context, plus documented Dual Chunk Attention and sparse-attention techniques for extending toward 1M tokens.[2]
- Open-weight Apache-2.0 release; the catalog exposes tool-calling and reasoning parameters through a standard API.[1][2]
Limitations
- Thinking mode is mandatory and the vendor notes an increased thinking length, so token use and latency will be higher than non-reasoning models; it is recommended mainly for highly complex tasks.[1][2]
- Performance evidence is limited to vendor-published tables and one aggregator catalog; independent verification is thin, so validate on your own workload before adopting.[1][2]
- Text-only input and output; served context may be capped below the native maximum (one catalog lists 128K), and self-hosting at extreme context is memory-intensive, with the card citing roughly 1 TB of GPU memory for 1M tokens.[1][2]
Where to use it
- Editorial suggestion: difficult math, science, and competition-style programming problems where extended step-by-step reasoning is useful.[2]
- Editorial suggestion: agentic pipelines that call external tools, for example via the vendor's Qwen-Agent framework or API tool parameters.[1][2]
- Editorial suggestion: analysis of very long documents or codebases, where the deployment budget can support large contexts.[2]
Sources
- [1] OpenRouter catalog · Qwen: Qwen3 235B A22B Thinking 2507Sources checked: Nov 13, 2026
- [2] Model documentation · Qwen: Qwen3 235B A22B Thinking 2507Sources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
Qwen3-235B-A22B-Thinking-2507 is a high-performance, open-weight Mixture-of-Experts (MoE) language model optimized for complex reasoning tasks. It activates 22B of its 235B parameters per forward pass and natively supports up to 262,144...
Specifications
- Model ID
- qwen/qwen3-235b-a22b-thinking-2507
- Creator
- qwen
- Added to OpenRouter
- 25.07.2025
- Context length
- 128,000
- Modality
- text->text
- Input modalities
- text
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.450 / out $3.5
- Knowledge cutoff
- 30.06.2025
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | No measurement | reasoning: mandatory · parameters: include_reasoning, reasoning |
| Coding | CC — below the B band | — |
| Agents | CC — below the B band | parameters: tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: mandatory · parameters: include_reasoning, reasoning |
| Document analysis | No measurement | — |
| Vision | No measurement | — |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
| Source | Benchmark | Value | Grade | Data retrieved |
|---|---|---|---|---|
| Artificial Analysis | Coding index | 22.1 | CC — below the B band | 13.11.2026 |
| Artificial Analysis | Agentic index | 1.3 | CC — below the B band | 13.11.2026 |
| Design Arena | models · 3dwin rate 40.4% · rank 120 | 980Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · codecategorieswin rate 40.8% · rank 126 | 1,023Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · datavizwin rate 32.6% · rank 124 | 950Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · gamedevwin rate 34.2% · rank 134 | 947Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · uicomponentwin rate 34% · rank 130 | 936Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · websitewin rate 42% · rank 125 | 1,045Elo | CC — below the B band | 13.11.2026 |
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| Venicefp8 | 128,000 | $0.450 | $3.5 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.